In all, 385 non-metastatic patients with histologically confirmed primary NPC were prospectively recruited between January 2011 and December 2013. The eligibility criteria are as follows: (1) histologically confirmed primary NPC; (2) no treatment before entering this study; (3) Eastern Cooperative Oncology Group (ECOG) status of 0 to 2; (4) age ≥ 18 years; (5) normal renal and hepatic function; (6) no evidence of distant metastases at initial diagnosis; and (7) complete pretreatment EBV DNA and tumor volume data. Patient groups are as follows: (1) lactation or pregnancy; (2) history of previous or synchronous secondary malignant tumors; (3) refusing treatment or quitting therapy during the process of receiving treatment; or (4) loss of the follow-up data were excluded. All patients were staged as stage II according to the seventh edition of the International Union against Cancer/American Joint Committee on Cancer staging system for NPC.

In our study, 198 patients were treated with IMRT alone, and the remaining 187 patients were treated with CCRT. Details of the radiotherapy techniques were reported in a previous study [18], with the dose of 68-70 Gy for the primary region and 60-68 Gy for metastatic cervical lymph nodes. For the 187 patients treated with CCRT, 142 patients received chemotherapy consisting of cisplatin given on the first, fourth, and seventh week of radiotherapy, and 45 patients received chemotherapy consisting of cisplatin given every week during radiotherapy.

Before treatment, blood samples were taken from each eligible patient for measurement of the plasma EBV DNA level. Blood samples were saved temporarily in EDTA-containing tubes until routinely measured by a real-time quantitative polymerase chain reaction. The detailed method was described in previous studies [19]. In this study, we divided the patients into a detectable group (> 0 copy/mL) and an undetectable group (0 copy/mL) according to the pretreatment plasma EBV DNA level.

At first, all patients received the computed tomography (CT) simulation scan (Plus 4, Siemens, Erlangen, Germany), including plain and enhanced CT, which extended from the top of head to the 2-cm region below the sub-clavicle head with a thickness of 3 mm. Then, contouring of the target region was determined by an IMRT planning system (NOMOS, Pittsburgh, PA) based on the institutional treatment protocol [20]. Our treatment plans were reviewed and approved by at least three radiation oncologists. Finally, tumor volume, regardless of gross tumor volume of nasopharynx (GTVnx) and gross tumor volume of cervical lymph node (GTVnd), was calculated by the IMRT planning system. GTVtotal was calculated by the addition of GTVnx and GTVnd (GTVtotal=GTVnx+GTVnd). In our study, regional lymph nodes with a shortest axial diameter of 11 mm in the jugulodigastric region, 5 mm in the retropharyngeal region and 10 mm in all the other regions of the neck were considered malignant. Besides, if there was a group of three or more lymph nodes with a borderline size, and those with necrosis or extracapsular spread in an irrespective size, these were also considered malignant.

The following baseline clinical data of all eligible patients were collected before treatment: sex, age, treatment, smoking status, concurrent diseases, pathological type, performance status grade evaluated by the ECOG, hereditary NPC, invasion of retropharyngeal lymph nodes, cervical lymph nodes and parapharyngeal space, viral capsid antigen (VCA)‒IgA, early antigen (EA)‒IgA [21,22], plasma EBV DNA level, GTVnx, and GTVnd.

After completion of their treatment, patients were evaluated at 3-month intervals for the first 3 years and then at 6-month intervals in the following years. In this study, our primary endpoint was PFS, OS, LRFS, and DMFS were our secondary endpoints.

PFS was calculated from the date of initial treatment of NPC to the date of first relapse at any site, to the date of death due to any cause, or to the date of the last follow-up visit if patients were still alive. OS was calculated from the date of initial treatment of NPC to the date of death due to any cause, or to the date of the last follow-up visit of the patient. LRFS and DMFS were calculated from the date of initial treatment of NPC to the date of locoregional relapse and distant relapse, respectively, or to the date of the last follow-up visit. Patients who were still alive on the last follow-up date were censored on the date of the last follow-up. Patients who lost contact in the process of following up were censored on the date of the last contact.

In our study, characteristics of the patients were described by the frequency and their corresponding percentages. Continuous variables in this study were described by median values if non-normal or mean (standard deviation), followed by their 25th-75th percentiles (interquartile range [IQR]). We also utilized the receiver operating characteristic (ROC) curve analysis to choose the optimum cutoff point of total tumor volume (GTVtotal)，GTVnx, and GTVnd. In brief, the specificity and sensitivity for the outcome being studied at each cutoff point were plotted to generate a ROC curve. The one localized closest to the point at both maximum specificity and sensitivity was chosen to be the optimum cutoff point. The Spearman correlation test was performed to evaluate the relationship between EBV DNA levels and GTVtotal, GTVnx, and GTVnd. The Mann-Whitney U test was utilized to compare the differences of the plasma EBV DNA level and that of GTVtotal in different subgroups separated by the involvement status of retropharyngeal lymph nodes, cervical lymph nodes and parapharyngeal space or clinical outcomes (locoregional failure, distant failure, or death). The Kaplan-Meier method and log-rank test were then used to evaluate the differences in survival probabilities for PFS, OS, LRFS, and DMFS between four subgroups divided according to the EBV DNA levels and tumor volume (details as follows). Potentially important prognostic factors (age, sex, treatment, smoking status, NPC history, involvement of parapharyngeal space, retropharyngeal lymph node and cervical lymph node, VCA-IgA, EA-IgA and combination of tumor volume and EBV DNA) were entered into multivariate analysis using Cox proportional hazards model. Finally, hazard ratios (HRs) as well as their corresponding 95% confidence intervals (CIs) were estimated by means of the Cox proportional hazards regression. All p-values quoted in this study were two-sided, and a p-value of ≤ 0.05 was considered to be statistically significant. Statistics analyses were all performed utilizing the IBM SPSS ver. 20.0 (IBM Corp., Armonk, NY).

This study was approved by the Clinical Research Ethics Committee of Sun Yat-sen University Cancer Center. Before treatment, written informed consent was obtained from each patient.

The characteristics of the 385 eligible patients with stage II NPC are listed in Table 1. Among these 385 NPC patients, 12 patients finally developed locoregional failure alone, 11 patients developed distant failure alone, one patient developed both locoregional and distant failures, and six patients died due to any cause. The median follow-up time was 49.87 months, with an IQR of 16.22 (42.01-58.23). The 3-year PFS rate, LRFS rate, DMFS rate, and OS rate were 93.5%, 96.6%, 96.9%, and 98.4%, respectively. The median value of the EBV DNA was 0 copy/mL, with a detection rate of 41.8%. The median values of the GTVtotal, GTVnx, and GTVnd were 25.70 cm³, 16.30 cm³, and 6.70 cm³, respectively.

Of the 385 eligible patients, the median EBV DNA levels of the patients with (n=25) or without (n=360) progression were 601.0 copies/mL (IQR, 0 to 16,450.0) and 0 copy/mL (IQR, 0 to 1,687.5), respectively (p=0.008). Median EBV DNA levels of the patients with (n=12) or without (n=373) distant failure were 2,545.0 copies/mL (IQR, 122.3 to 20,875.0) and 0 copy/mL (IQR, 0 to 1,935.0), respectively (p=0.013). For GTVtotal, the median values with (n=25) or without (n=360) progression were 35.20 cm³ (IQR, 22.75 to 50.35) and 24.60 cm³ (IQR, 16.80 to 35.95), respectively (p=0.015). Median values with (n=13) or without (n=372) locoregional failure were 38.50 cm³ (IQR, 31.65 to 55.65) and 24.70 cm³ (IQR, 16.80 to 36.38), respectively (p=0.002).

The detectable rates of pretreatment plasma EBV DNA for patients with or without involvement of retropharyngeal lymph nodes, cervical lymph nodes and parapharyngeal space were 46.8% vs. 32.6%, 54.9% vs. 28.4%, and 46.7% vs. 38.1%, respectively. The median EBV DNA level and GTVtotal of patients with involvement of retropharyngeal lymph nodes were 0 copy/mL (IQR, 0 to 4,105.0) and 27.8 cm³ (IQR, 19.2 to 40.9), respectively, compared with 0 copy/mL (IQR, 0 to 513.0, p=0.003) and 21.6 cm³ (IQR, 15.2 to 32.4, p < 0.001) for patients without involvement of retropharyngeal lymph nodes. Those of patients with involvement of cervical lymph nodes were 345.0 copies/mL (IQR, 0 to 6,710.0) and 30.6 cm³ (IQR, 19.6 to 43.9), compared with 0 copy/mL (IQR, 0 to 94.0, p < 0.001) and 23.1 cm³ (IQR, 15.4 to 31.1; p < 0.001) for patients without metastasis of cervical lymph nodes. GTVtotal for patients with or without involvement of parapharyngeal space was 29.6 cm³ (IQR, 20.6 to 44.0) and 23.2 cm³ (IQR, 15.4 to 34.4; p < 0.001), respectively. Detailed values are listed in Table 2.

As described above, the optimum cutoff point of tumor volume was determined by the ROC curve analysis. As presented in S1A Fig., because our primary endpoint was PFS in this study, the optimum cutoff point of tumor volume for PFS was 29.35 cm³. In addition, for potential acceptance and application in clinical procedures, we chose the nearest integer, 30 cm³, as the optimum cutoff point in this study. Similarly, optimum cutoff points of GTVnx and GTVnd were 20 cm³ and 10 cm³, respectively (S1B and S1C Fig.).

By means of Spearman correlation analysis, we found that both GTVtotal and GTVnd were correlated with EBV DNA, with the correlation coefficients of 0.360 (p < 0.001) and 0.379 (p < 0.001), respectively. However, the GTVnx was weakly correlated with EBV DNA, with the correlation coefficient of 0.101 (p=0.047).

In this study, we found that patients with detectable EBV DNA (> 0 copy/mL) had shorter PFS, LRFS, and DMFS compared with those with undetectable EBV DNA (0 copy/mL). The 3-year PFS, LRFS, and DMFS rates for the detectable and undetectable EBV DNA groups were 89.1% vs. 96.4%, 94.3% vs. 98.2%, and 94.2% vs. 98.6%, with corresponding p-values of 0.005, 0.039, and 0.017, respectively (Fig. 1A, S2A and S2B Fig.). However, there was no difference in 3-year OS rate between detectable and undetectable EBV DNA groups (p=0.769) (S2C Fig.). Similarly, we also noted that patients with a larger GTVtotal (≥ 30 cm³) had shorter PFS and LRFS compared with those patients with a smaller GTVtotal (< 30 cm³). The 3-year PFS and LRFS rates for the larger and smaller GTVtotal groups were 89.2% vs. 96.1% and 92.6% vs. 99.1%, respectively, with the corresponding p-values of 0.008 and 0.001 (Fig. 1B, S3A Fig.). There was also no difference in 3-year DMFS (p=0.431) (S3B Fig.) and OS rate (p=0.620) (S3C Fig.) between the larger and smaller GTVtotal groups. Similar phenomena were observed for GTVnx and GTVnd as that of tumor volume (Fig. 1C and D, S4 and S5 Figs.).

According to the results described as above, both EBV DNA and GTVtotal are effective prognostic factors for PFS, LRFS, and DMFS, but not for OS. Therefore, we divided the entire population into four groups according to the cutoff points of pretreatment EBV DNA level and GTVtotal mentioned above: the undetectable EBV DNA (0 copy/mL) and smaller GTVtotal (< 30 cm³) group (group A), the undetectable EBV DNA and larger GTVtotal (≥ 30 cm³) group (group B), the detectable EBV DNA (> 0 copy/mL) and smaller GTVtotal group (group C), and the detectable EBV DNA and larger GTVtotal group (group D). Interestingly, we found that there was no statistical significance for 3-year PFS and DMFS rates between group A and group B, groups B and C, and groups A and C. Therefore, we further divided the patients into the high-risk group (EBV DNA > 0 copy/mL, GTVtotal ≥ 30 cm³) and low-risk group (EBV DNA 0 copy/mL, GTVtotal < 30 cm³; EBV DNA 0 copy/mL, GTVtotal ≥ 30 cm³; or EBV DNA > 0 copy/mL, GTVtotal < 30 cm³). Differences for PFS, LRFS, and DMFS (Fig. 2) were statistically significant between the low-risk group and the high-risk group. The 3-year PFS, LRFS, and DMFS rates between the low-risk and high-risk groups were 95.5% vs. 86.8% (p=0.003), 97.9% vs. 92.3% (p=0.010), and 97.9% vs. 93.4% (p=0.031), respectively.

In our study, Kaplan-Meier method and log-rank test were used to evaluate the differences between the low-risk group and high-risk group in patients treated with radiotherapy (RT) (n=198) alone and in patients treated with CCRT (n=187). Differences between the low-risk group and highrisk group in patients treated with RT alone were not statistically significant for PFS, with 3-year PFS rates of 95.9% vs. 87.0% (p=0.064). Similarly, differences between the low-risk group and high-risk group in patients treated with CCRT were also not statistically significant for PFS, with 3-year PFS rates of 94.8% vs. 86.7% (p=0.055). Detailed information is shown in Fig. 3.

The results of multivariate analysis are listed in Table 3. In this study, we found that only the combination of tumor volume and pretreatment plasma EBV DNA level (low-risk group vs. high-risk group) reached a p-value of ≤ 0.05, with a corresponding HR value of 2.804 (95% CI, 1.113 to 7.064).